\begin{table}[!h]
\centering
\caption{Effect of Regime Type on Framing by Attribute\label{tab:regimereg}}
\centering
\begin{threeparttable}
\begin{tabular}[t]{llll}
\toprule
\multicolumn{1}{c}{\em{ }} & \multicolumn{3}{c}{\em{Similarity with Country Dictionary (\%)}} \\
  & Chaos (1) & Corruption (2) & Sports (3)\\
\midrule
Government Accounts & -0.11 & -7.18*** & -3.19***\\
 & (1.05) & (0.98) & (0.74)\\
Democracy (vdem) & -0.42 & 1.40 & 5.59\\
 & (7.71) & (8.70) & (5.98)\\
Govt. Acct. x Democracy & 2.79*** & 3.86*** & 0.15\\
 & (0.96) & (1.02) & (0.82)\\
ln Frequency 1 & -1.58*** & -1.96*** & -2.28***\\
 & (0.16) & (0.16) & (0.16)\\
ln Frequency 2 & -3.57*** & -2.31*** & -1.71***\\
 & (0.09) & (0.06) & (0.06)\\
\midrule
\addlinespace[]
\multicolumn{4}{l}{\textit{Statistics}}\\
\hspace{1em}Observations & 93453 & 48636 & 97180\\
\addlinespace[]
\multicolumn{4}{l}{\textit{Fixed effects}}\\
\hspace{1em}Subscription Account & Yes & Yes & Yes\\
\hspace{1em}Year & Yes & Yes & Yes\\
\hspace{1em}Dictionary & Yes & Yes & Yes\\
\bottomrule
\end{tabular}
\begin{tablenotes}
\item Notes: $^{*}$p$<$0.1; $^{**}$p$<$0.05; $^{***}$p$<$0.01 White robust standard-errors clustered on country in parentheses. Dependent variable is cosine similarity between country and respective attribute dictionaries. Controls for dictionary term frequency are shown. Unit of analysis is the object-attribute word pair, which varies according to the object and attribute dictionary size.
\end{tablenotes}
\end{threeparttable}
\end{table}
